A Sleep and Circadian Biomarker-Based Predictive Model for Differentiating Unipolar and Bipolar Depression

Differentiating unipolar depression (UDD) from bipolar depression (BDD) remains a major clinical challenge with important treatment implications. Sleep-related markers, both subjective (such as hypersomnia in BDD and insomnia in UDD) and objective (actigraphy, polysomnography [PSG]), show promise, yet no study has integrated these approaches into a single predictive model.

Patients with DSM-5-TR-defined UDD or BDD in a depressive episode underwent clinical, questionnaire, actigraphy, and PSG assessments. Discriminating variables were entered into a backward stepwise logistic regression (BSLR) to derive the optimal classification model.

The study drew on 159 patients: 43 with BDD (42 actigraphy, 20 PSG) and 116 with UDD (93 actigraphy, 44 PSG). Compared to BDD, patients with UDD reported poorer sleep quality (Pittsburgh sleep quality index [PSQI]), more severe insomnia (insomnia severity index [ISI]), and lower sleep efficiency (SE, actigraphy). Patients with BDD showed longer total rest time per 24 h, lower average activity during the least active 5-h period (L5) and the 10 most active hours (M10), and higher N2% and total NREM sleep (PSG). Six variables were retained in the final BSLR model, explaining 49.4% of the variance. The most discriminative were higher PSQI/ISI and lower actigraphic SE in UDD, versus longer rest time, lower L5 activity, and higher N2% in BDD. The model showed excellent discriminative ability (AUC = 0.926, sensitivity = 0.882, specificity = 0.895, Youden's index = 0.777), with strong predictive values (positive predictive value [PPV] = 88.3%, negative predictive value [NPV] = 89.5%).

Integrating subjective, actigraphic, and polysomnographic markers provides a promising, noninvasive approach to differentiate UDD from BDD.